[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN111657918B - Falling detection method and system combining electrocardio and inertial sensing data - Google Patents

Falling detection method and system combining electrocardio and inertial sensing data Download PDF

Info

Publication number
CN111657918B
CN111657918B CN202010537555.5A CN202010537555A CN111657918B CN 111657918 B CN111657918 B CN 111657918B CN 202010537555 A CN202010537555 A CN 202010537555A CN 111657918 B CN111657918 B CN 111657918B
Authority
CN
China
Prior art keywords
data
fall
inertial sensing
ecg
daily
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN202010537555.5A
Other languages
Chinese (zh)
Other versions
CN111657918A (en
Inventor
刘勇国
陶文元
杨尚明
李巧勤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010537555.5A priority Critical patent/CN111657918B/en
Publication of CN111657918A publication Critical patent/CN111657918A/en
Application granted granted Critical
Publication of CN111657918B publication Critical patent/CN111657918B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1116—Determining posture transitions
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1116—Determining posture transitions
    • A61B5/1117—Fall detection
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2562/00—Details of sensors; Constructional details of sensor housings or probes; Accessories for sensors
    • A61B2562/02—Details of sensors specially adapted for in-vivo measurements
    • A61B2562/0219—Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Physiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Molecular Biology (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Dentistry (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention discloses a method and a system for fall detection by combining electrocardio and inertial sensing data, wherein the method comprises the following steps: collecting fall data and daily motion data of a human body in the daily activity process by utilizing equipment comprising an ECG (electrocardiogram) and an inertial sensing unit; dividing the collected fall data and daily motion data according to the same time period, extracting to obtain inertial sensing data characteristics and ECG data characteristics, and performing normalization processing to obtain normalized sample data characteristics; performing dimensionality reduction on the normalized sample data features by adopting a principal component analysis algorithm to obtain dimensionality reduction data features, and classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model; and inputting the newly acquired data into a support vector machine classification model to obtain a falling detection result. The change of the human body falling process is reflected from two different dimensions of the physiological activity and the posture of the human body respectively, the defect of inaccurate single type data is avoided, and the falling detection precision is improved.

Description

Falling detection method and system combining electrocardio and inertial sensing data
Technical Field
The invention relates to the field of human body fall detection, in particular to a fall detection method and system combining electrocardio and inertial sensing data.
Background
The falling causes psychological and physical injuries of the old people, and is one of the important factors threatening the health of the old people. The fall detection means that human body activity data are collected, and a fall incident is found through data analysis. The equipment for collecting the human activity data is divided into environment sensing equipment and wearable equipment. The environment sensing equipment is not required to be worn by a human body, but the monitoring range is limited. Wearing equipment is mainly an inertial sensor, and the equipment is small in size, can be worn on one's body, and is widely applied in the field of fall detection, but the accuracy is influenced by the wearing position.
In recent years there has been research into the use of wearable Electrocardiogram (ECG) monitors for detecting falls. The ECG signal represents the bioelectricity activity caused by the periodic contraction of the heart and the relaxation of the cardiac muscle, and can reflect the change of the heart activity of the human body in different postures and posture conversion processes. Chinese patent "CN 201510697530.0 a method and system for fall detection based on electrocardiogram" provides a method for fall detection based on heart rate. The method comprises the steps that air pressure and heart rate data are acquired through a detection terminal, firstly, height change of a human body is judged through the air pressure, if the height change reaches a threshold value, heart rate detection is started, and if a heart rate change value is larger than the threshold value, falling is judged. The wearable equipment is used for observing the ECG morphological change degree of the human body in lying and standing postures for fall detection, and the accuracy rate is 77.30%.
Disclosure of Invention
The invention aims to solve the technical problems that in the existing detection technology, the falling detection based on the environment sensing equipment has privacy and limited detection range, the falling detection precision based on the inertia sensing equipment is influenced by the wearing position, the falling detection based on the electrocardiogram mainly detects the change of the physiological state of a human body by whether the heart rate reaches a threshold value, the electrocardiogram information is not fully utilized, and the detection precision is influenced.
The invention is realized by the following technical scheme:
a method of fall detection incorporating electrocardiographic and inertial sensing data, comprising:
s1: the method comprises the steps that a wrist wearable device comprising an ECG and an inertial sensing unit is used for collecting falling data and daily action data of a human body in a daily activity process;
s2: dividing the collected fall data and daily action data according to the same time period;
s3: respectively extracting the features of the divided fall data and daily motion data to obtain inertial sensing data features and ECG data features;
s4: carrying out normalization processing on the inertial sensing data characteristics and the ECG data characteristics to obtain normalized sample data characteristics;
s5: performing dimensionality reduction on the normalized sample data features by adopting a principal component analysis algorithm to obtain dimensionality reduction data features, and classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
s6: and inputting the newly acquired daily activity data of the wrist wearable device into the support vector machine classification model to obtain a falling detection result.
According to the invention, different daily activity data, namely falling data and normal activity data, are collected, falling actions including forward slipping, backward slipping, lateral slipping, tripping and stunning and normal daily activities including walking, running, going up and down stairs, posture conversion and bending are simulated to obtain the support vector machine classification model, and the support vector machine classification model is classified into falling or non-falling, so that the input new daily activity data is input into the support vector machine classification model to obtain a falling detection result, namely falling or non-falling.
The wearable wrist device comprising the ECG, namely the wearable electrocardiogram detector and the inertial sensor is used for collecting different types of data in daily activities, the ECG data are different in data in different activity states, the defects that the detection range is limited when only the inertial sensor is used for detection and the falling detection precision is influenced by the wearing position to cause inaccurate falling detection results can be overcome, the wearable wrist device and the inertial sensor are combined, the change of the falling process of a human body is reflected from two different dimensions of the physiological activity and the posture of the human body respectively, the defect that single type of data is inaccurate is avoided, more accurate falling data are obtained, and the falling detection precision is improved.
The daily activities of the present invention include falls and normal daily activities. Wherein the falling action is simulated on the soft pad and comprises forward slipping, backward slipping, lateral slipping, tripping and stunning; the normal daily activities include walking, running, going up and down stairs, posture conversion, stooping and the like.
The types of the ECG data and the inertial sensing data are completely different, and the data obtained by the two data in normal activities and in falling are different, the value ranges are different and the dimensions are different, so that all the data need to be segmented in the same time period, effective inertial sensing data and effective ECG data need to be extracted, linear normalization processing is adopted for the inertial sensing data and the ECG data, and all the data are mapped to 0-1; the dimentional expression is transformed into the dimensionless expression by adopting the normalization function to become a scalar, so that the calculation process can be simplified, and the calculation time of the detection process can be reduced; in order to improve the fall detection accuracy to a greater extent, it is necessary to extract various features from the inertial sensing data, including: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy; a normal cycle of ECG signal in ECG acquired data consists of QRS wave, T wave and P wave, where the P wave corresponds to the left and right atrial depolarization phases, and the frequency spectrum ranges from 10-15 Hz. The QRS complex corresponds to left and right ventricular depolarizations in the frequency spectrum ranging from 10-40 Hz. The T wave reflects the ventricular repolarization process, in the range of around 300 milliseconds after QRS compounding. It is therefore necessary to perform QRS complex determination first, followed by P and T wave determination for each cardiac cycle of the ECG signal.
Further, the S3 includes:
s31: extracting inertial sensing data of the divided falling data and daily action data in the inertial sensing unit to obtain inertial sensing data characteristics;
the inertial sensing data characteristics include: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy;
s32: extracting data characteristics of a QRS complex, a P wave and a T wave in the ECG data of the segmented falling data and daily motion data to obtain the ECG data characteristics;
the normalization processing function in S4:
Figure GDA0002618947210000031
where max represents the maximum value of the sample data and min represents the minimum value of the sample data. x represents raw sensor data and x represents the normalized result.
Further, the extraction of the QRS complex:
and extracting a peak point of the QRS complex according to the peak value of the ECG data, extracting the lowest points before and after the peak point respectively, and determining the initial position and the offset of the QRS complex to obtain the QRS complex.
Further, the following time difference equation is used for filtering to obtain ynT：
ynT＝-2*χ(n-2)T-χ(n-1)T+χ(n+1)T+2*χ(n+2)T
Wherein, χnTRepresents a data sample of size n at time T;
the extraction formula of the P wave is as follows:
θ＝0.30*(μn-μHF)
where θ represents a dynamic threshold representation of ECG data for a certain time period, μnMeans the slope average, mu, of all data points in the most recent signal time seriesHFHigh frequency noise in one bounce;
wherein,
Figure GDA0002618947210000032
μHFreferring to high frequency noise in one trip, to calculate the high frequency noise, the signal is first passed to a high pass filter:
y′nT＝χnT-2*χ(n-1)T+χ(n-2)T
dividing the ith sample high-pass filtered signal into k segments, respectively denoted as y'nT(i,1),y′nT(i,2),…,y′nT(i, k) the average of the high pass filtered signal over all segments is calculated as
Figure GDA0002618947210000033
The high frequency noise present in the cardiac cycle can be estimated as follows:
Figure GDA0002618947210000034
wherein R isAIs the amplitude of the R peak, K is the amplification factor obtained by the experiment;
and when theta reaches a threshold value, extracting the peak value of the ECG data of the time segment, extracting the lowest value point of the slope according to the slopes of the waveform data before and after the peak value through a time sequence, and determining the initial position and the offset of the P wave to obtain the P wave.
Further, the T wave extraction formula is:
θ＝0.30*(μn-μ′HF)
where θ represents a dynamic threshold representation of ECG data for a certain time period, μnRefers to the slope average, μ 'of all data points in the most recent signal time series'HFLow frequency noise in one beat;
μ′HFreferring to low-frequency noise in one-time jump, in order to calculate the low-frequency noise, firstly, smoothing is carried out through a 2-order low-pass filter:
y″nT＝2*y(n-1)T-y(n-2)T+χnT-2*χ(n-4)T+χn-8)T
the signal of the ith sample after low-pass filtering is divided into k segments, which are respectively denoted as y ″nT(i,1),y″nT(i,2),…,y″nT(i, k) the average of the low pass filtered signals over all segments is calculated as
Figure GDA0002618947210000041
The low frequency noise present in the cardiac cycle can be estimated as follows:
Figure GDA0002618947210000042
wherein R isAIs the amplitude of the R peak, K is the amplification factor obtained by the experiment;
and when theta reaches a threshold value, after a time derivative is adopted to extract a peak value of the ECG data of the time period, determining the initial position and the offset of the T wave according to the minimum curvature radius positions before and after the peak value through a time sequence order, and obtaining the T wave.
Further, the S5 includes:
s51: marking the normalized sample data characteristics to obtain fall sample data and daily activity sample data;
s52: performing dimensionality reduction on the fall sample data and the daily activity sample data by adopting a principal component analysis algorithm to obtain dimensionality reduction data characteristics;
the principal component analysis algorithm:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix;
s53: classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
given a training data set D { (x)i,yi)|xi∈R,yi∈{-1,1}}
Wherein x isiIs the inertial sensing unit and the electrocardiogram feature vector after dimensionality reduction, each xiContaining k-dimensional features, yiFor the fall category, the value of this scheme is 1 or-1 (corresponding to falls and non-falls). Two types of linearly classified hyperplanes can be defined as: for yi＝1，wTxi+ b.gtoreq.1 for yi＝-1，wTxi+ b ≦ -1, where w is the weight vector and b is the offset. The goal is to reduce the distance w of two hyperplanes to the maximum, which can be expressed as a quadratic optimization problem:
Figure GDA0002618947210000051
the decision function can separate two categories (fall and non-fall):
f(xi)＝sign(wTxi+b)
wherein, f (x)i) Class, x, representing data characteristicsiRepresenting a characteristic vector of the dimension reduction data, w representing a weighting vector, and b representing a deviation;
for all xiWhen f (x)i) When the number is 1, the detection category is falling; when f (x)i) When the value is-1, the detection category is daily movement.
Further, the principal component analysis algorithm comprises the steps of:
forming n rows and m columns of feature matrix X by the fall sample data and the daily activity sample data according to columns;
carrying out zero equalization operation on each row of elements of the feature matrix X;
calculating covariance between different features in the feature matrix X to obtain a covariance matrix, wherein the covariance calculation formula is as follows:
Figure GDA0002618947210000052
wherein
Figure GDA0002618947210000053
Xj (k)Representing fall sample data and feature data X in daily activity samplesi，XjValues in the kth fall sample data and the daily activity sample data; while
Figure GDA0002618947210000054
Then the feature data X in all fall sample data and daily activity samples is representediThe average value of (a) of (b),
Figure GDA0002618947210000055
then the feature data X in all fall sample data and daily activity samples is representedjN represents each fall sample data and daily activity sample dataA characteristic dimension of (d);
calculating an eigenvalue of the covariance matrix and a corresponding eigenvector;
arranging the eigenvectors into a matrix from top to bottom according to the size of the corresponding eigenvalue, and taking the first k rows to form an eigenvector matrix P;
classifying fall detection according to fall sample data and daily activity sample data of different dimensions, determining the value of k, and obtaining the dimension reduction data characteristics:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix.
Further, the inertial sensing unit comprises a three-axis accelerometer and a three-axis screw instrument.
Further, the fall data and daily movement data comprise ECG data, acceleration data and spirometer data.
A fall detection system incorporating electrocardiographic and inertial sensing data, comprising:
the acquisition module comprises an ECG and an inertial sensing unit and is used for acquiring falling data and daily action data of a human body in the daily activity process;
the data processing module is used for extracting the features of the segmented falling data and daily motion data to obtain inertial sensing data features and ECG data features, and performing normalization processing on the extracted inertial sensing data features and ECG data features;
the model construction module is used for carrying out dimensionality reduction processing on the normalized sample data characteristics to obtain dimensionality reduction data characteristics, and classifying the dimensionality reduction data characteristics by adopting a support vector machine algorithm to obtain a support vector machine classification model;
and the falling detection module is used for inputting the newly acquired daily activity data of the wrist wearable equipment into the support vector machine classification model to obtain a falling detection result.
Further, the inertial sensing data characteristics include: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy;
the ECG data includes QRS complex, P-wave and T-wave data features.
Compared with the prior art, the invention has the following advantages and beneficial effects:
according to the method and the system for detecting the falling of the human body, provided by the invention, the ECG and the inertial sensing data are combined, the change of the falling process of the human body is reflected from two different dimensions, namely the physiological activity and the posture of the human body respectively, the information is more comprehensive, the defect of single type of data is avoided, and the falling detection accuracy is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a schematic overall flow diagram of the present invention;
FIG. 2 is another flow chart of the present invention;
FIG. 3 is a schematic representation of ECG data according to the present invention;
FIG. 4 is a schematic diagram of the system of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
Example 1
As shown in fig. 1 and 2, a fall detection method combining electrocardiographic and inertial sensing data includes:
s1: the method comprises the steps that a wrist wearable device comprising an ECG and an inertial sensing unit is used for collecting falling data and daily action data of a human body in a daily activity process;
s2: dividing the collected fall data and daily action data according to the same time period;
s3: respectively extracting the features of the divided fall data and daily motion data to obtain inertial sensing data features and ECG data features;
s4: carrying out normalization processing on the inertial sensing data characteristics and the ECG data characteristics to obtain normalized sample data characteristics;
s5: performing dimensionality reduction on the normalized sample data features by adopting a principal component analysis algorithm to obtain dimensionality reduction data features, and classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
s6: and inputting the newly acquired daily activity data of the wrist wearable device into the support vector machine classification model to obtain a falling detection result.
The daily activities of the present invention include falls and normal daily activities. Wherein the falling action is simulated on the soft pad and comprises forward slipping, backward slipping, lateral slipping, tripping and stunning; the normal daily activities include walking, running, going up and down stairs, posture conversion, stooping and the like.
Further, the S3 includes:
s31: extracting inertial sensing data of the divided falling data and daily action data in the inertial sensing unit to obtain inertial sensing data characteristics;
the inertial sensing data characteristics include: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy;
the inertial sensing unit comprises two sensing units, namely a three-axis accelerometer and a three-axis screw instrument, wherein the falling data and the daily action data of the daily activity process of the human body comprise two types of acceleration and screw instrument data, and nine characteristics are extracted from each type of data;
therefore, a total of 54 inertial sensor data features, where 3 denotes that there are three axes for acceleration and angular velocity of the gyroscope, are extracted, and the inertial sensor data needs to be extracted for each axis of acceleration and angular velocity of the gyroscope, and therefore needs to be multiplied by 3.
S32: extracting data characteristics of a QRS complex, a P wave and a T wave in the ECG data of the segmented falling data and daily motion data to obtain the ECG data characteristics;
one normal cycle of the ECG signal consists of the QRS wave, the T wave and the P wave, as shown in FIG. 3. The P wave corresponds to the depolarization phase of the left and right atria, with a frequency spectrum in the range of 10-15 Hz. The QRS complex corresponds to left and right ventricular depolarizations in the frequency spectrum ranging from 10-40 Hz. The T wave reflects the ventricular repolarization process, in the range of around 300 milliseconds after QRS compounding.
The normalization processing function in S4:
Figure GDA0002618947210000071
where max represents the maximum value of the sample data and min represents the minimum value of the sample data. x represents raw sensor data and x represents the normalized result.
Further, the extraction of the QRS complex:
and extracting a peak point of the QRS complex according to the peak value of the ECG data, extracting the lowest points before and after the peak point respectively, and determining the initial position and the offset of the QRS complex to obtain the QRS complex.
Further, the following time difference equation is used for filtering to obtain ynT：
ynT＝-2*χ(n-2)T-χ(n-1)T+χ(n+1)T+2*χ(n+2)T
Wherein, χnTRepresents a data sample of size n at time T;
the extraction formula of the P wave is as follows:
θ＝0.30*(μn-μHF)
where θ represents a dynamic threshold representation of ECG data for a certain time period, μnMeans the slope average, mu, of all data points in the most recent signal time seriesHFHigh frequency noise in one bounce;
wherein
Figure GDA0002618947210000081
μHFReferring to high frequency noise in one trip, to calculate the high frequency noise, the signal is first passed to a high pass filter:
y′nT＝χnT-2*χ(n-1)T+χ(n-2)T
dividing the ith sample high-pass filtered signal into k segments, respectively denoted as y'nT(i,1),y′nT(i,2),…,y′nT(i, k) the average of the high pass filtered signal over all segments is calculated as
Figure GDA0002618947210000082
The high frequency noise present in the cardiac cycle can be estimated as follows:
Figure GDA0002618947210000083
wherein R isAIs the amplitude of the R peak, K is an amplification factor obtained by experiment, and this embodiment is set to 40;
and when theta reaches a threshold value, extracting the peak value of the ECG data of the time segment, extracting the lowest value point of the slope according to the slopes of the waveform data before and after the peak value through a time sequence, and determining the initial position and the offset of the P wave to obtain the P wave.
Further, the T wave extraction formula is: θ is 0.30 ═ μ (μ)n-μ′HF)
Where θ represents a dynamic threshold representation of ECG data for a certain time period, μnRefers to the slope average, μ 'of all data points in the most recent signal time series'HFLow frequency noise in one beat;
μ′HFreferring to low-frequency noise in one-time jump, in order to calculate the low-frequency noise, firstly, smoothing is carried out through a 2-order low-pass filter: y ″)nT＝2*y(n-1)T-y(n-2)T+χnT-2*χ(n-4)T+χ(n-8)T
The signal of the ith sample after low-pass filtering is divided into k segments, which are respectively denoted as y ″nT(i,1),y″nT(i,2),…,y″nT(i, k) the average of the low pass filtered signals over all segments is calculated as
Figure GDA0002618947210000084
The low frequency noise present in the cardiac cycle can be estimated as follows:
Figure GDA0002618947210000085
wherein R isAIs the amplitude of the R peak, K is the amplification factor obtained by the experiment;
and when theta reaches a threshold value, after a time derivative is adopted to extract a peak value of the ECG data of the time period, determining the initial position and the offset of the T wave according to the minimum curvature radius positions before and after the peak value through a time sequence order, and obtaining the T wave.
The present embodiment extracts 14 features from the ECG data through the above steps, as shown in table 1;
TABLE 1 ECG Signal characteristics
Serial number Feature(s)
1 P wave duration (millisecond)
2 QRS wave duration (millisecond)
3 Duration of Q wave (millisecond)
4 Duration of R wave (millisecond)
5 Duration of S wave (millisecond)
6 Duration of T wave (millisecond)
7 QRS wave deflection (millisecond)
8 P wave peak amplitude (μ V)
9 QRS peak amplitude (μ V)
10 Q wave amplitude (μ V)
11 R wave amplitude (μ V)
12 S wave amplitude (mu V)
13 T wave peak amplitude (μ V)
14 QRS wave area (μ V)
The S5 includes:
s51: marking the normalized sample data characteristics to obtain fall sample data and daily activity sample data;
s52: performing dimensionality reduction on the fall sample data and the daily activity sample data by adopting a principal component analysis algorithm to obtain dimensionality reduction data characteristics;
the principal component analysis algorithm:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix;
s53: classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
given a training data set D { (x)i,yi)|xi∈R,yi∈{-1,1}}
Wherein x isiIs the inertial sensing unit and the electrocardiogram feature vector after dimensionality reduction, each xiContaining k-dimensional features, yiFor the fall category, the value of this scheme is 1 or-1 (corresponding to falls and non-falls). Two types of linearly classified hyperplanes can be defined as: for yi＝1，wTxi+ b.gtoreq.1 for yi＝-1，wTxi+ b ≦ -1, where w is the weight vector and b is the offset. The goal is to reduce the distance w of two hyperplanes to the maximum, which can be expressed as a quadratic optimization problem:
Figure GDA0002618947210000101
the decision function can separate two categories (fall and non-fall):
f(xi)＝sign(wTxi+b)
wherein, f (x)i) Class, x, representing data characteristicsiRepresenting a characteristic vector of the dimension reduction data, w representing a weighting vector, and b representing a deviation;
for all xiWhen f (x)i) When the number is 1, the detection category is falling; when f (x)i) When the value is-1, the detection category is daily movement.
Further, the principal component analysis algorithm comprises the steps of:
setting m 68-dimensional data, and forming a feature matrix X with 68 rows and m columns by the falling sample data and the daily activity sample data according to columns;
carrying out zero equalization operation on each row of elements of the feature matrix X;
calculating covariance between different features in the feature matrix X to obtain a covariance matrix, wherein the covariance calculation formula is as follows:
Figure GDA0002618947210000102
wherein
Figure GDA0002618947210000103
Xj (k)Representing fall sample data and feature data X in daily activity samplesi，XjValues in the kth fall sample data and the daily activity sample data; while
Figure GDA0002618947210000104
Then the feature data X in all fall sample data and daily activity samples is representediThe average value of (a) of (b),
Figure GDA0002618947210000105
then the feature data X in all fall sample data and daily activity samples is representedjN represents the characteristic dimensions of each fall sample data and daily activity sample data;
calculating an eigenvalue of the covariance matrix and a corresponding eigenvector;
arranging the eigenvectors into a matrix from top to bottom according to the size of the corresponding eigenvalue, and taking the first k rows to form an eigenvector matrix P;
classifying fall detection according to fall sample data and daily activity sample data of different dimensions, determining the value of k, and obtaining the dimension reduction data characteristics:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix.
Further, the inertial sensing unit comprises a three-axis accelerometer and a three-axis screw instrument.
Further, the fall data and daily movement data comprise ECG data, acceleration data and spirometer data.
As shown in fig. 4, a fall detection system combining electrocardiographic and inertial sensing data comprises:
the acquisition module comprises an ECG and an inertial sensing unit and is used for acquiring falling data and daily action data of a human body in the daily activity process;
the data processing module is used for extracting the features of the segmented falling data and daily motion data to obtain inertial sensing data features and ECG data features, and performing normalization processing on the extracted inertial sensing data features and ECG data features;
the model construction module is used for carrying out dimensionality reduction processing on the normalized sample data characteristics to obtain dimensionality reduction data characteristics, and classifying the dimensionality reduction data characteristics by adopting a support vector machine algorithm to obtain a support vector machine classification model;
and the falling detection module is used for inputting the newly acquired daily activity data of the wrist wearable equipment into the support vector machine classification model to obtain a falling detection result.
Further, the inertial sensing data characteristics include: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy;
the ECG data includes QRS complex, P-wave and T-wave data features.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (8)

1. A fall detection method combining electrocardiogram and inertial sensing data is characterized by comprising the following steps:
s1: the method comprises the steps that a wrist wearable device comprising an ECG and an inertial sensing unit is used for collecting falling data and daily action data of a human body in a daily activity process;
s2: dividing the collected fall data and daily action data according to the same time period;
s3: respectively extracting the features of the divided fall data and daily motion data to obtain inertial sensing data features and ECG data features;
the S3 includes:
s31: extracting inertial sensing data of the divided falling data and daily action data in the inertial sensing unit to obtain inertial sensing data characteristics;
the inertial sensing data characteristics include: arithmetic mean, standard deviation, median absolute deviation, maximum value, minimum value, frequency signal deflection, frequency signal kurtosis, maximum frequency component, average energy;
s32: extracting data characteristics of a QRS complex, a P wave and a T wave in the ECG data of the segmented falling data and daily motion data to obtain the ECG data characteristics;
the extraction formula of the P wave is as follows:
θ＝0.30*(μn-μHF)
where θ represents a dynamic threshold representation of ECG data for a certain time period, μnMeans the slope average, mu, of all data points in the most recent signal time seriesHFHigh frequency noise in one bounce;
wherein, a certain period of time means: a time period containing a P-wave;
the latest signal time series refers to: currently calculated munA section of signal formed by corresponding n data points;
the one-time jumping means: corresponding to one ECG segmentation cycle;
when theta reaches a threshold value, extracting a peak value of the ECG data of the time period, extracting a slope minimum value point according to slopes of waveform data before and after the peak value through a time sequence order, and determining a P wave initial position and an offset to obtain a P wave;
s4: carrying out normalization processing on the inertial sensing data characteristics and the ECG data characteristics to obtain normalized sample data characteristics;
s5: performing dimensionality reduction on the normalized sample data features by adopting a principal component analysis algorithm to obtain dimensionality reduction data features, and classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
s6: and inputting the newly acquired daily activity data of the wrist wearable device into the support vector machine classification model to obtain a falling detection result.
2. A fall detection method combining electrocardiographic and inertial sensing data according to claim 1,
the normalization processing function in S4:
Figure FDA0003116728730000021
where max represents the maximum sample data value, min represents the minimum sample data value, x represents the raw sensor data, and x represents the normalization result.
3. A fall detection method combining electrocardiographic and inertial sensing data according to claim 2, characterized in that the extraction of the QRS complex:
and extracting a peak point of the QRS complex according to the peak value of the ECG data, extracting the lowest points before and after the peak point respectively, and determining the initial position and the offset of the QRS complex to obtain the QRS complex.
4. A fall detection method combining electrocardiographic and inertial sensing data according to claim 2,
the T wave extraction formula:
θ＝0.30*(μn-μ′HF)
where θ represents a dynamic threshold representation of ECG data for a certain time period, μnRefers to the slope average, μ 'of all data points in the most recent signal time series'HFLow frequency noise in one beat;
and when theta reaches a threshold value, after a time derivative is adopted to extract a peak value of the ECG data of the time period, determining the initial position and the offset of the T wave according to the minimum curvature radius positions before and after the peak value through a time sequence order, and obtaining the T wave.
5. A method for fall detection with combined electrocardiographic and inertial sensing data according to claim 1, wherein the S5 comprises:
s51: marking the normalized sample data characteristics to obtain fall sample data and daily activity sample data;
s52: performing dimensionality reduction on the fall sample data and the daily activity sample data by adopting a principal component analysis algorithm to obtain dimensionality reduction data characteristics;
the principal component analysis algorithm:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix;
s53: classifying the dimensionality reduction data features by adopting a support vector machine algorithm to obtain a support vector machine classification model;
the support vector machine algorithm:
f(xi)＝sign(wTxi+b)
wherein, f (x)i) Class, x, representing data characteristicsiRepresenting the feature vector of the dimension reduction data, w representing the weighting vector and b representing the deviation.
6. A fall detection method combining electrocardiographic and inertial sensing data according to claim 5, characterized in that the principal component analysis algorithm step:
forming n rows and m columns of feature matrix X by the fall sample data and the daily activity sample data according to columns;
carrying out zero equalization operation on each row of elements of the feature matrix X;
calculating covariance between different features in the feature matrix X to obtain a covariance matrix, wherein the covariance calculation formula is as follows:
Figure FDA0003116728730000031
wherein
Figure FDA0003116728730000032
Representing fall sample data and feature data X in daily activity samplesi，XjValues in the kth fall sample data and daily activity sample data, and
Figure FDA0003116728730000033
then the feature data X in all fall sample data and daily activity samples is representediThe average value of (a) of (b),
Figure FDA0003116728730000034
then the feature data X in all fall sample data and daily activity samples is representedjN represents the characteristic dimensions of each fall sample data and daily activity sample data;
calculating an eigenvalue of the covariance matrix and a corresponding eigenvector;
arranging the eigenvectors into a matrix from top to bottom according to the size of the corresponding eigenvalue, and taking the first k rows to form an eigenvector matrix P;
classifying fall detection according to fall sample data and daily activity sample data of different dimensions, determining the value of k, and obtaining the dimension reduction data characteristics:
Y＝PX
wherein Y represents the dimensionality reduction data characteristic, P represents a characteristic vector matrix, and X represents a characteristic matrix.
7. A fall detection method combining electrocardiographic and inertial sensing data according to claim 1, wherein the inertial sensing unit comprises a three-axis accelerometer and a three-axis gyroscope.
8. A method of fall detection combining electrocardiographic and inertial sensing data according to claim 1 wherein the fall data and daily motion data comprise ECG data, acceleration data and gyroscope data.
CN202010537555.5A 2020-06-12 2020-06-12 Falling detection method and system combining electrocardio and inertial sensing data Expired - Fee Related CN111657918B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010537555.5A CN111657918B (en) 2020-06-12 2020-06-12 Falling detection method and system combining electrocardio and inertial sensing data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010537555.5A CN111657918B (en) 2020-06-12 2020-06-12 Falling detection method and system combining electrocardio and inertial sensing data

Publications (2)

Publication Number Publication Date
CN111657918A CN111657918A (en) 2020-09-15
CN111657918B true CN111657918B (en) 2021-09-24

Family

ID=72387602

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010537555.5A Expired - Fee Related CN111657918B (en) 2020-06-12 2020-06-12 Falling detection method and system combining electrocardio and inertial sensing data

Country Status (1)

Country Link
CN (1) CN111657918B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115530800A (en) * 2021-06-30 2022-12-30 中国科学技术大学 Falling detection system based on bracelet

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106725495A (en) * 2017-01-13 2017-05-31 深圳先进技术研究院 A kind of fall detection method, apparatus and system
CN109171738A (en) * 2018-07-13 2019-01-11 杭州电子科技大学 Fall detection method based on human body acceleration multiple features fusion and KNN
CN109691994A (en) * 2019-01-31 2019-04-30 英菲泰克(天津)科技有限公司 A kind of rhythm of the heart analysis method based on electrocardiogram
CN110946573A (en) * 2019-11-01 2020-04-03 东软集团股份有限公司 Cardiac arrest detection device, detection model training device, method and equipment

Family Cites Families (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4184493A (en) * 1975-09-30 1980-01-22 Mieczyslaw Mirowski Circuit for monitoring a heart and for effecting cardioversion of a needy heart
US6070097A (en) * 1998-12-30 2000-05-30 General Electric Company Method for generating a gating signal for cardiac MRI
US7218960B1 (en) * 2003-06-24 2007-05-15 Pacesetter, Inc. System and method for detecting cardiac ischemia based on T-waves using an implantable medical device
US9943244B2 (en) * 2007-09-14 2018-04-17 Angel Medical Systems, Inc. Waveform feature value averaging system and methods for the detection of cardiac events
CN104771178B (en) * 2015-04-13 2018-01-05 深圳市飞马与星月科技研究有限公司 The method and device of identification
CN105869354B (en) * 2016-04-29 2017-12-01 华南理工大学 A kind of Falls Among Old People detection method based on attractor propagation algorithm
CN105796112B (en) * 2016-05-11 2018-10-30 江苏财经职业技术学院 Falls Among Old People detects monitor system
CN106407996A (en) * 2016-06-30 2017-02-15 华南理工大学 Machine learning based detection method and detection system for the fall of the old
CN108354610A (en) * 2017-08-29 2018-08-03 浙江好络维医疗技术有限公司 A kind of Falls Among Old People detection method and detecting system based on three-axis sensor and EGC sensor
CN108257352B (en) * 2017-12-30 2019-12-20 广州柏颐信息科技有限公司 Fall detection and early warning method based on intelligent wearable equipment
CN108245172B (en) * 2018-01-10 2020-04-07 山东大学 Human body posture recognition method free of position constraint
CN108498091B (en) * 2018-05-08 2021-08-31 复旦大学附属中山医院 Two-electrode electrocardiogram lead intelligent positioning system
CA3065208C (en) * 2018-12-20 2024-06-18 Queen's University At Kingston Long qt syndrome diagnosis and classification
CN110353694B (en) * 2019-07-18 2022-04-05 哈尔滨工业大学 Motion recognition method based on feature selection
CN110659595A (en) * 2019-09-10 2020-01-07 电子科技大学 Tumble type and injury part detection method based on feature classification
CN110742599B (en) * 2019-11-01 2022-05-10 广东工业大学 Electrocardiosignal feature extraction and classification method and system
CN111199187A (en) * 2019-12-11 2020-05-26 中国科学院计算机网络信息中心 Animal behavior identification method based on algorithm, corresponding storage medium and electronic device
CN111603170A (en) * 2020-06-01 2020-09-01 河南华南医电科技有限公司 Human body position detection method based on vector cardiogram

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106725495A (en) * 2017-01-13 2017-05-31 深圳先进技术研究院 A kind of fall detection method, apparatus and system
CN109171738A (en) * 2018-07-13 2019-01-11 杭州电子科技大学 Fall detection method based on human body acceleration multiple features fusion and KNN
CN109691994A (en) * 2019-01-31 2019-04-30 英菲泰克(天津)科技有限公司 A kind of rhythm of the heart analysis method based on electrocardiogram
CN110946573A (en) * 2019-11-01 2020-04-03 东软集团股份有限公司 Cardiac arrest detection device, detection model training device, method and equipment

Also Published As

Publication number Publication date
CN111657918A (en) 2020-09-15

Similar Documents

Publication Publication Date Title
EP3065628B1 (en) Biomechanical activity monitoring
Tapia et al. Real-time recognition of physical activities and their intensities using wireless accelerometers and a heart rate monitor
CN102138789B (en) Dynamic electrocardiogram and motion recording and analyzing system
CN105769173B (en) A kind of cardioelectric monitor system with electrocardiosignal noise removal function
Khan et al. Accelerometer's position free human activity recognition using a hierarchical recognition model
CN202027572U (en) Dynamic recording and analyzing device for electrocardiogram and movement
US20190320944A1 (en) Biomechanical activity monitoring
Fan et al. Wearable motion attitude detection and data analysis based on Internet of Things
CN112263244A (en) Gait-based fatigue degree evaluation system and method
Zhang et al. Upper limb motion capturing and classification for unsupervised stroke rehabilitation
CN105796091B (en) A kind of intelligent terminal for removing electrocardiosignal vehicle movement noise
CN113609975A (en) Modeling method for tremor detection, hand tremor detection device and method
CN111657918B (en) Falling detection method and system combining electrocardio and inertial sensing data
Pan et al. Evaluation of hemiplegic gait based on plantar pressure and inertial sensors
Erin et al. Spectral Analysis of Cardiogenic Vibrations to Distinguish Between Valvular Heart Diseases.
Krassnig et al. User-friendly system for recognition of activities with an accelerometer
CN114469041A (en) Heart rate change data characteristic analysis method in exercise process
Ding-Fei et al. Study of feature extraction based on autoregressive modeling in EGG automatic diagnosis
Pawar et al. Impact of ambulation in wearable-ECG
Nouredanesh et al. Detection of compensatory balance responses using wearable electromyography sensors for fall-risk assessment
CN111297366B (en) Data processing method and diagnosis device for assisting disease diagnosis based on daily necessities
KR20070087787A (en) Emotion classification system of real time bio signals and emotion classification method
CN116999057A (en) Hemiplegia gait recognition and hemiplegia gait evaluation method based on wearable sensor
Alpoim et al. Human activity recognition systems: state of art
Ahamed et al. Using wavelet-based fractal analysis of inertial measurement unit signals to examine gait data from men and women during a load carriage task

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210924

CF01 Termination of patent right due to non-payment of annual fee